def __init__(self,
city_name,
avoid_stopping,
memory_fraction=0.25,
image_cut=[115, 510]):
#Agent.__init__(self)
dir_path = os.path.dirname(__file__)
#self.dropout_vec = [1.0] * 8 + [0.7] * 2 + [0.5] * 2 + [0.5] * 1 + [0.5, 1.] *6
self.dropout_vec = [1.0] * 8 + [0.7] * 2 + [0.5] * 2 + [0.5] * 1 + [
0.5, 1.
] * 6
self._image_size = (88, 200, 3)
self._avoid_stopping = True
self._image_cut = image_cut
tf.reset_default_graph()
config_gpu = tf.ConfigProto(allow_soft_placement=True)
# GPU to be selected, just take zero , select GPU with CUDA_VISIBLE_DEVICES
config_gpu.gpu_options.visible_device_list = '0'
config_gpu.gpu_options.per_process_gpu_memory_fraction = memory_fraction
self._sess = tf.Session(config=config_gpu)
self._models_path = "/home/pankaj/Trainer_module/CARLAILtrainer/models/" #"D:/outbox/changed_old_trainer/trainer5/models/"#dir_path + '/model/'
self._sess.run(tf.global_variables_initializer())
#self.load_model()
with tf.device('/gpu:0'):
saver = tf.train.import_meta_graph(self._models_path +
'model.ckpt.meta')
self._graph = tf.get_default_graph()
self._input_images = self._graph.get_tensor_by_name('input_image:0')
self._input_speed = self._graph.get_tensor_by_name('input_speed:0')
self._dout = self._graph.get_tensor_by_name('dropout:0')
self._follow_lane = self._graph.get_tensor_by_name(
'Network/Branch_0/fc_8:0')
self._left = self._graph.get_tensor_by_name('Network/Branch_1/fc_11:0')
self._right = self._graph.get_tensor_by_name(
'Network/Branch_2/fc_14:0')
self._straight = self._graph.get_tensor_by_name(
'Network/Branch_3/fc_17:0')
self._speed = self._graph.get_tensor_by_name(
'Network/Branch_4/fc_20:0')
self._intent = self._graph.get_tensor_by_name(
'Network/Branch_5/fc_23:0')
with tf.device('/gpu:0'):
saver.restore(self._sess, self._models_path + 'model.ckpt')
self._curr_dir = 0
self.count = 0
self._enable_manual_control = False
self._is_on_reverse = False
#pygame.init()
#self._display = pygame.display.set_mode(
# (WINDOW_WIDTH, WINDOW_HEIGHT),
# pygame.HWSURFACE | pygame.DOUBLEBUF)
self.command_follower = CommandFollower(city_name)
self.traffic_light_infraction = False
def make_controlling_agent(args, town_name):
"""
Make the controlling agent object depending on what was selected.
Options:
Forward Agent: Trivial agent that just accelerate forward.
Human Agent: Agent controlled by a human driver, currently only by keyboard.
Command Follower: A* planner followed by a PID controller (used as expert for data collection)
"""
if args.controlling_agent == "ForwardAgent":
return ForwardAgent()
elif args.controlling_agent == "HumanAgent":
return HumanAgent()
elif args.controlling_agent == "CommandFollower":
return CommandFollower(town_name)
else:
raise ValueError("Selected Agent Does not exist")
def make_controlling_agent(args, town_name):
""" Make the controlling agent object depending on what was selected.
Right now we have the following options:
Forward Agent: A trivial agent that just accelerate forward.
Human Agent: An agent controlled by a human driver, currently only by keyboard.
"""
if args.controlling_agent == "ForwardAgent":
return ForwardAgent()
elif args.controlling_agent == "HumanAgent":
# TDNextPR: Add parameters such as joysticks to the human agent.
return HumanAgent()
elif args.controlling_agent == "CommandFollower":
return CommandFollower(town_name)
elif args.controlling_agent == 'LaneFollower':
return LaneFollower(town_name)
else:
raise ValueError("Selected Agent Does not exist")
class ImitationLearning(Agent):
def __init__(self,
city_name,
avoid_stopping,
memory_fraction=0.25,
image_cut=[115, 510]):
#Agent.__init__(self)
dir_path = os.path.dirname(__file__)
#self.dropout_vec = [1.0] * 8 + [0.7] * 2 + [0.5] * 2 + [0.5] * 1 + [0.5, 1.] *6
self.dropout_vec = [1.0] * 8 + [0.7] * 2 + [0.5] * 2 + [0.5] * 1 + [
0.5, 1.
] * 6
self._image_size = (88, 200, 3)
self._avoid_stopping = True
self._image_cut = image_cut
tf.reset_default_graph()
config_gpu = tf.ConfigProto(allow_soft_placement=True)
# GPU to be selected, just take zero , select GPU with CUDA_VISIBLE_DEVICES
config_gpu.gpu_options.visible_device_list = '0'
config_gpu.gpu_options.per_process_gpu_memory_fraction = memory_fraction
self._sess = tf.Session(config=config_gpu)
self._models_path = "/home/pankaj/Trainer_module/CARLAILtrainer/models/" #"D:/outbox/changed_old_trainer/trainer5/models/"#dir_path + '/model/'
self._sess.run(tf.global_variables_initializer())
#self.load_model()
with tf.device('/gpu:0'):
saver = tf.train.import_meta_graph(self._models_path +
'model.ckpt.meta')
self._graph = tf.get_default_graph()
self._input_images = self._graph.get_tensor_by_name('input_image:0')
self._input_speed = self._graph.get_tensor_by_name('input_speed:0')
self._dout = self._graph.get_tensor_by_name('dropout:0')
self._follow_lane = self._graph.get_tensor_by_name(
'Network/Branch_0/fc_8:0')
self._left = self._graph.get_tensor_by_name('Network/Branch_1/fc_11:0')
self._right = self._graph.get_tensor_by_name(
'Network/Branch_2/fc_14:0')
self._straight = self._graph.get_tensor_by_name(
'Network/Branch_3/fc_17:0')
self._speed = self._graph.get_tensor_by_name(
'Network/Branch_4/fc_20:0')
self._intent = self._graph.get_tensor_by_name(
'Network/Branch_5/fc_23:0')
with tf.device('/gpu:0'):
saver.restore(self._sess, self._models_path + 'model.ckpt')
self._curr_dir = 0
self.count = 0
self._enable_manual_control = False
self._is_on_reverse = False
#pygame.init()
#self._display = pygame.display.set_mode(
# (WINDOW_WIDTH, WINDOW_HEIGHT),
# pygame.HWSURFACE | pygame.DOUBLEBUF)
self.command_follower = CommandFollower(city_name)
self.traffic_light_infraction = False
#self._initialize_game()
#def load_model(self):
def get_keyboard_control(self, keys):
"""
Return a VehicleControl message based on the pressed keys. Return None
if a new episode was requested.
"""
if keys[K_r]:
return None
control = Control()
if keys[K_LEFT] or keys[K_a]:
control.steer = -1.0
if keys[K_RIGHT] or keys[K_d]:
control.steer = 1.0
if keys[K_UP] or keys[K_w]:
control.throttle = 1.0
if keys[K_DOWN] or keys[K_s]:
control.brake = 1.0
if keys[K_SPACE]:
control.hand_brake = True
if keys[K_q]:
self._is_on_reverse = not self._is_on_reverse
if keys[K_p]:
self._enable_manual_control = not self._enable_manual_control
control.reverse = self._is_on_reverse
return control
def run_step(self, measurements, sensor_data, directions, target):
self.state = self.command_follower.run_step(measurements, sensor_data,
directions, target)
#pygame.event.get()
#key_control = self.get_keyboard_control(pygame.key.get_pressed())
if self._enable_manual_control:
self.state.update({
"predicted_stop_pedestrian": 1,
"predicted_stop_vehicle": 1,
"predicted_stop_traffic": 1
})
return key_control, self.state
else:
model_control = self._compute_action(
sensor_data['CameraRGB'].data,
measurements.player_measurements.forward_speed, directions)
return model_control, self.state
def _compute_action(self, rgb_image, speed, direction=None):
rgb_image = rgb_image[self._image_cut[0]:self._image_cut[1], :]
image_input = scipy.misc.imresize(
rgb_image, [self._image_size[0], self._image_size[1]])
image_input = np.multiply(image_input, 1.0 / 255.0)
steer, acc, brake, pred_intents, pred_speed = self._control_function(
image_input, speed, direction)
#self.command_follower.param_controller['target_speed'] = 30.0 # Fixed speed limit to 30 km/hr for testing
if acc > brake:
brake = 0.0
control = self.command_follower.controller.get_control(
self.state['wp_angle'], self.state['wp_angle_speed'],
min(pred_intents), speed * 3.6, steer, acc, brake)
#control = Control()
#control.steer = steer
#control.throttle = acc
#control.brake = brake
control.hand_brake = 0
control.reverse = 0
ped_intent, tra_intent, veh_intent = pred_intents
self.state.update({
"predicted_stop_pedestrian": ped_intent,
"predicted_stop_vehicle": veh_intent,
"predicted_stop_traffic": tra_intent
})
print(
f"Controls: Steer: {control.steer:.2f} \tThrottle: {control.throttle:.2f} \tBrake: {control.brake:.2f} Command: {numtoCommands[direction]} "
)
print(
f" True ped_intent: {self.state['stop_pedestrian']}\t True veh_intent: {self.state['stop_vehicle']} \t True tra_intent: {self.state['stop_traffic_lights']}"
)
print(
f"Pred_ped_intent: {ped_intent} \tPred_veh_intent: {veh_intent} \tPred_tra_intent: {tra_intent} "
)
return control
def _control_function(self, image_input, speed, control_input):
image_input = image_input.reshape(
(1, self._image_size[0], self._image_size[1], self._image_size[2]))
# Normalize with the maximum speed from the training set ( 90 km/h)
curr_speed = speed
speed = np.array(float(speed) / MAX_SPEED)
speed = speed.reshape((1, 1))
print(f"Current Speed: ", speed * 3.6 * MAX_SPEED)
if control_input == 2 or control_input == 0.0:
branch = self._follow_lane
elif control_input == 3:
branch = self._left
elif control_input == 4:
branch = self._right
else:
branch = self._straight
feedDict = {
self._input_images: image_input,
self._input_speed: speed,
self._dout: [1] * len(self.dropout_vec)
}
output, predicted_speed, pred_intent = self._sess.run(
[branch, self._speed, self._intent], feed_dict=feedDict)
predicted_steers, predicted_acc, predicted_brake = output[0]
predicted_speed = predicted_speed[0]
print(f"Predicted Speed: ", predicted_speed * 3.6 * MAX_SPEED)
for i in range(len(pred_intent[0])):
if pred_intent[0][i] > 0.9:
pred_intent[0][i] = 1
elif pred_intent[0][i] <= 0.05:
pred_intent[0][i] = 0
os.system('clear')
if (predicted_speed * 3.6 * MAX_SPEED) > 5 and (
speed * 3.6 * MAX_SPEED) < 5: #False braking
print("False braking !!!!")
predicted_acc = 0.5
predicted_brake = 0
#if (tra_intent - self.state['stop_traffic_lights']) > 0.7:
# print(self.state['stop_traffic_lights'], tra_intent)
# self.traffic_light_infraction = True
return predicted_steers, predicted_acc, predicted_brake, pred_intent[
0], predicted_speed[0] * MAX_SPEED * 3.6